This model serves as the baseline for the Aerial Wildfire Suppression environment, trained and tested on task 2 with difficulty 1 using the Proximal Policy Optimization (PPO) algorithm.
Environment: Aerial Wildfire Suppression
Task: 2
Difficulty: 1
Algorithm: PPO
Episode Length: 3000
Training max_steps: 1800000
Testing max_steps: 180000
Train & Test Scripts
Download the Environment
Evaluation results
- Crash Count on hivex-aerial-wildfire-suppressionself-reported0.10833333656191826 +/- 0.13545462085140364
- Extinguishing Trees on hivex-aerial-wildfire-suppressionself-reported12.764999979734421 +/- 20.904079059070074
- Extinguishing Trees Reward on hivex-aerial-wildfire-suppressionself-reported63.82500011920929 +/- 104.52039570112963
- Fire Out on hivex-aerial-wildfire-suppressionself-reported0.46333333626389506 +/- 0.3753984235292453
- Fire too Close to City on hivex-aerial-wildfire-suppressionself-reported0.7550000011920929 +/- 0.37763111996880777
- Preparing Trees on hivex-aerial-wildfire-suppressionself-reported561.5350036621094 +/- 325.15177018123586
- Preparing Trees Reward on hivex-aerial-wildfire-suppressionself-reported2807.674984741211 +/- 1625.7588563088673
- Water Drop on hivex-aerial-wildfire-suppressionself-reported19.77666656970978 +/- 9.682256516548964
- Water Pickup on hivex-aerial-wildfire-suppressionself-reported19.191666650772095 +/- 9.573244118078943
- Cumulative Reward on hivex-aerial-wildfire-suppressionself-reported3287.165838623047 +/- 1376.9107350360173